Image segmentation method, system and electronic device based on depth learning

Abstract

The invention belongs to the technical field of image segmentation, in particular to an image segmentation method, a system and an electronic device based on depth learning. The image segmentation method based on depth learning comprises the following steps: step A: normalizing the original image; B, inputting the normalized image into a ResUNet network model, extracting a feature map containing global semantic information from the input image by the ResUNet network model, upsampling the feature map and stacking the feature map to obtain a final feature map; step c: classifying pixel by pixelthe feature map after upsampling and stacking processing, and outputting the image segmentation result. The present application solves the common problem of gradient disappearance in a convolution neural network, and also enables the network to be more easily trained and converged to a better segmentation result.

Description

technical field [0001] The present application belongs to the technical field of image segmentation, and in particular relates to an image segmentation method, system and electronic equipment based on deep learning. Background technique [0002] Image engineering is mainly divided into three levels: image processing, image analysis and image understanding. The goal of image segmentation is to classify each pixel in the image, that is, to use the discontinuity between the pixel of the image target object and the surrounding pixels or the similarity within the target to cut the picture into multiple small-area segmentation blocks. Image segmentation is the most basic and difficult part of image engineering, and it is an important bridge between image processing and image analysis. Image segmentation methods mainly include binary segmentation, semantic segmentation and instance segmentation. In the traditional image processing algorithms, most of them study binary segmentatio...

AI Technical Summary

Problems solved by technology

[0009] Although the deep neural network has far surpassed the traditional segmentation method, there are still some problems in image segmentation using CNN (Convolutional Neural Network, Convolutional Neural Network), such as the reduction of feature resolution and the existence of multi-scale objects, which cannot be well Model global context information, etc.

Although U-Net uses the method of downsampling-upsampling, too much spatial detail information of the picture will be lost in the process of downsampling

DRN uses atrous convolution to ensure that the resolution of the smallest feature map will not be too small, but it only alleviates the problem of feature resolution reduction. If the resolution does not decrease, the video memory of the graphics card cannot accommodate too many feature maps of the image. The feature diversity of the image will be reduced

PSPNet uses the pyramid pooling method to obtain global context information, the effect is not bad, but the edge of the segmented picture is still rough

Images